nVidia caught cheating on benchmarks

[Kuciwalker]

Originally posted by Azazel
Imran

Sorry, Fez. I think you should take it in good humor. We're not laughing AT you, you know.

actually we are, but it's too funny to pass up

[Asher]

Originally posted by Fez
Hey Ashboy, do something with that cute boyfriend of yours... I am tonight with my SO.. he is eating Chinese take out food here in my room..

I've gone out with him like 5 times in the past week, he's still in high school and he's got homework to do, so I'm not bugging him for the next few days (we're talking on MSN though).

Besides, I work from 8am-9pm tomorrow. :/

[Giancarlo]

Originally posted by Asher

I've gone out with him like 5 times in the past week, he's still in high school and he's got homework to do, so I'm not bugging him for the next few days (we're talking on MSN though).

Besides, I work from 8am-9pm tomorrow. :/

Well we will agree on one thing.. and the fact he is really handsome. Well I wish you luck with that. Sure I might not like your CPU choice.. but you got a nice guy there...

I just finished up high school speaking of that... Gotta do my graduation on the 6th and 7th of June.

[tinyp3nis]

Uhhh. Nothing wrong with optimizing? Depends how you do it. If you really want to know what's going on, here is some links.
http://www.extremetech.com/article2/...1086857,00.asp
http://www.tomshardware.com/technews...23_192553.html
JC's take on drivers
http://slashdot.org/comments.pl?sid=65617&cid=6051216

Basically, what ATI did, is acceptable and "everybody does it". Even so, ATI is removing their optimization from their next catalyst.

But unlike Asher said, what Nvidia did, isn't what everybody does and is not accepted, well here is a good qoute from the 3rd link: "Rewriting shaders behind an application's back in a way that changes the output under non-controlled circumstances is absolutely, positively wrong and indefensible."
Also it doesn't cost houndreds of thousands dollars to participate in Futuremarks beta program, 5000$ is the minimum.

[Kuciwalker]

I'm not sure I understand what you (plural) mean by "optimization". In the normal sense, it's a good thing, but from the context it seems like it means "optimization towards the benchmark, without necessarily improving actual performance". Right?

[Asher]

Nvidia was part of Futuremark's program but pulled out when Futuremark decided to heavily slant 3DMark2003 towards PS 1.4, which is rather ridiculous seeing as only 2 games in existance use them. Everything else uses 1.1 or 1.3. By coincidence, Nvidia has to emulate 1.4 so it's much slower.

And ATI has done the same thing. Am I the only one that remembers quack?

[Asher]

Originally posted by skywalker
I'm not sure I understand what you (plural) mean by "optimization". In the normal sense, it's a good thing, but from the context it seems like it means "optimization towards the benchmark, without necessarily improving actual performance". Right?

Because it's optimized for the benchmark, it does improve actual performance. Not for 3DMark2003, mind you, since it's not an actual game. But frames per second do go up with the optimization.

ATI's been busted for it, Nvidia's been busted for it. Nvidia actually replaces some shading programs because Futuremark (rather stupidly) wrote them in such a way so they're essentially tailor-made for the Radeon, with 3 pixels per pass (which is completely nonstandard, and I can't think of any game which uses it in the real world, and it doubles the amount of passes needed for Nvidia cards).

The fact is, 3DMark2003 is somewhat of a disaster.

It's about as reliable as Sysmark is now. Sysmark is biased towards Intel, 3DMark towards ATi.

[tinyp3nis]

Originally posted by skywalker
I'm not sure I understand what you (plural) mean by "optimization". In the normal sense, it's a good thing, but from the context it seems like it means "optimization towards the benchmark, without necessarily improving actual performance". Right?

Exactly. Just read the links if you have time and it's pretty clear.
Edit: Carmack said that he things how ATI handled 3dmarks GT4 is much more accepted way than how NVidia did it.

[tinyp3nis]

Originally posted by Asher
Nvidia was part of Futuremark's program but pulled out when Futuremark decided to heavily slant 3DMark2003 towards PS 1.4, which is rather ridiculous seeing as only 2 games in existance use them. Everything else uses 1.1 or 1.3. By coincidence, Nvidia has to emulate 1.4 so it's much slower.

And ATI has done the same thing. Am I the only one that remembers quack?

Everybody remembers it. It's the stuff that gets still thrown around when people don't know what the hell is going on right "now"(any new nvidia vs ati thing) and want to share their 2 cents without knowing anything that's actually happening. Only mention if it makes any point please.


Because of everyone remembering, and the fact that ATI hardware is pretty solid right now, they don't need reasons to cheat. Unlike Nvidia.

[Ned]

Fez, Ghost Recon's AI is the same. AI soldiers take and use cover. They also appear to use snipers or its equivalent, because if they spot you, wham, your dead.

There are aspects of GR that I have not seen in other games, at least not as real as in this game. If you have played MOHAA, remember the landing at Normandy that was a replay of Saving Private Ryan? Well, the best part of that mission was the tremendous incoming fire that keep you entirely rattled while you tried to figure out what to do next. Well, Ghost Recon takes that to the next level. The bullets really wizz around your head and make noise the loudness of which depends upon how close the impact is. Leaves in bushes and trees hit by the bullets kick and fall. Dust sprays up and bullet holes appear next to you. Many times, you cannot even see or hear what is firing at you. It is amazing. Real combat must be like this.

Also, many missions have objectives that must be met within a certain time. This forces you to quickly assault a "dug-in" defense or defend against an overwhelming assault while also dealing with meeting the objective.

There are no "health" restores in Ghost Recon missions. If a team member gets wounded, he or she stays that way during the mission. This is a problem because wounded team members at times move very slowly, complicating the mission.

Another major difference between Ghost Recon and other "team" FPS is that in Ghost Recon, you can (and must) play any and all of the members of your team on each mission. If one of your team members gets killed, they are not available for the next mission. This is critical since you have a limited number of troops from which to select a team to begin with, and each successful mission for a trooper allows you to increase his or her combat effectiveness.

Needless to say, I have replayed many missions from the beginning just to get through them with mininum casualties even with a very adequate quick save and quick restore capability. If you play on the "Elite" level without using quick save and quick restore, you really have to think and move fast to survive.

[Giancarlo]

I aways played MOHAA.. my favorite game indeed... I never played Ghost recon though.

In Vietcong, you usually have a medic. Or sometimes first aid kits. If you play on easy you can always get healed to 100%.. but if you play on normal or harder.. you will notice the medic only can do so much.. you sustain injury overtime no matter if you get treated by the medic.

That is exactly how it is in Vietcong too.. if you play Vietcong mode.. there is no save. And it is insanely difficult. I tried it. They say in the manual.. that is how it was like.

Also you have to keep your team alive because each member has a specific function.

[Asher]

Originally posted by tinyp3nis
Because of everyone remembering, and the fact that ATI hardware is pretty solid right now, they don't need reasons to cheat. Unlike Nvidia.

As far as I'm concerned, ATI paying off and sending free "developer assistants" to Futuremark to help develop 3DMark03 is just as much "cheating" as optimizing your drivers specifically for the benchmark...

Hell, ATI wrote the shader code for the troll-physics benchy.

[tinyp3nis]

Originally posted by Asher

As far as I'm concerned, ATI paying off and sending free "developer assistants" to Futuremark to help develop 3DMark03 is just as much "cheating" as optimizing your drivers specifically for the benchmark...

Hell, ATI wrote the shader code for the troll-physics benchy.

I'm going to need more than just a motive to do bad things. It may seem that they had opportunity, but did they really? Was there foul play, or just speculation?

[Asher]

Originally posted by tinyp3nis
I'm going to need more than just a motive to do bad things. It may seem that they had opportunity, but did they really? Was there foul play, or just speculation?

Nvidia and ATI were both in that developer beta program, but Nvidia left a long time ago in protest over how 3DMark03 was being written for ATI cards in mind, and have so much stuff be emulated on the Nvidia architecture. It makes even less sense, since the 1.1/1.3 pixel shaders are almost universally used in DX8 games, not ATI's 1.4.

[Ned]

Originally posted by Fez
I aways played MOHAA.. my favorite game indeed... I never played Ghost recon though.

In Vietcong, you usually have a medic. Or sometimes first aid kits. If you play on easy you can always get healed to 100%.. but if you play on normal or harder.. you will notice the medic only can do so much.. you sustain injury overtime no matter if you get treated by the medic.

That is exactly how it is in Vietcong too.. if you play Vietcong mode.. there is no save. And it is insanely difficult. I tried it. They say in the manual.. that is how it was like.

Also you have to keep your team alive because each member has a specific function.

Well, I don't even want to imagine what it would be like to be a Vietcong going up against the US Army and Marine Corps in Vietnam. We had an awesome firepower advantage because we had airmobile resupply of ammo, artillery, tactical air and B52s. The Vietcong had none of this. I think our "kill" ratio was at least 10-1 and more like 20-1.

(BTW, and this probably deserves its own thread, has anyone ever noticed that Russia took enormous casualties in assualting the German army in 44-45, suffering far more casualties than it inflicted on the Germans, while at the same time the US Army took very few casualties and inflicted many times that number on the Germans? Was the primary difference in the two theaters the use of airpower by the US? Or was there something more, like bad Russian tactics?)

[Giancarlo]

Originally posted by Ned


Well, I don't even want to imagine what it would be like to be a Vietcong going up against the US Army and Marine Corps in Vietnam. We had an awesome firepower advantage because we had airmobile resupply of ammo, artillery, tactical air and B52s. The Vietcong had none of this. I think our "kill" ratio was at least 10-1 and more like 20-1.

I would wonder.. the poor VC were pretty diehard in their beliefs.... so they fought to the death.

[tinyp3nis]

Originally posted by Asher

Nvidia and ATI were both in that developer beta program, but Nvidia left a long time ago in protest over how 3DMark03 was being written for ATI cards in mind, and have so much stuff be emulated on the Nvidia architecture. It makes even less sense, since the 1.1/1.3 pixel shaders are almost universally used in DX8 games, not ATI's 1.4.

Yes, it was about 1.4 going down to 1.1 skipping 1.3 if the card doesn't support 1.4? Or something like that?
I just took their word when they said that 1.3 has no real performance over 1.1. Nobody (well I didn't see it anywhere) challenged that argument so I have to go with what they(futuremark) said.
What goes for using 1.4 when the card supports it, I don't see any reason _not to use it_ if is perfoming better like they said and the card supports it. There is no reason why they should have gone how the games uptill then had done, because when released it was ment for the future games. Is there any reason to believe games won't support 1.4 from now on (now there are more cards that can do 1.4)?
There was a lot of discussion about this back then, and the answers sure satisfied me.

What I don't know is (I am not pro, just someone who reads news!):
What is more likely, games will use ps2.0 and go down to 1.1/1.3 in the future if the card doesnt support it (is this practical/possible)???
Or 1.4 and go down to 1.1/1.3?
Or just use 1.1/1.3.
It's about goddamn time games start using dx8/9, untill then my gf4 will stay in my comp... we'll have to wait.
As long as games don't use any cool shaders 3dmark2003 will not serve its purpose. The dawm of cinemating, where the hell are you, "I have been ready" for ages now.

[Asher]

Originally posted by tinyp3nis
Yes, it was about 1.4 going down to 1.1 skipping 1.3 if the card doesn't support 1.4? Or something like that?
I just took their word when they said that 1.3 has no real performance over 1.1. Nobody (well I didn't see it anywhere) challenged that argument so I have to go with what they(futuremark) said.

That's not the problem, though.

The GeForce FX supports 1.4, it just needs to be emulated. 3DMark pings the drivers, determines it has 1.4 support, and runs them. They're far slower than Nvidia's equivalent 1.3. One of the things they got busted for cheating is forcing the GeForce FXs to run it in 1.3, since it makes so much more sense.

What goes for using 1.4 when the card supports it, I don't see any reason _not to use it_ if is perfoming better like they said and the card supports it.

The only card it performs better on is the Radeon. Don't let the numbering confuse you, 1.4 is not inherently better than 1.3. They're just different versions that MS okayed. Nvidia made 1.1-1.3 (1.0 never got released), ATI made 1.4. They're just different ways to go about doing pixel shading.

Since ATI cards also support 1.1-1.3, virtually no game has implemented 1.4 since it's rather silly to. Which is why I'm completely dumbfounded why Futuremark made 3DMark03 so heavy on it. Hell, in all but one of the DX9 tests, it still uses 1.4 instead of 2.0! It's like somebody went out of their way to make the Radeon's performance better, but that's just my opinion. Either that, or gross stupidity.

What is more likely, games will use ps2.0 and go down to 1.1/1.3 in the future if the card doesnt support it (is this practical/possible)???
Or 1.4 and go down to 1.1/1.3?
Or just use 1.1/1.3.

What's most likely is no games, at all, will use 1.4. If it's a DX8 game, it'll use 1.3, if it's a DX9 game, it'll use 2.0.

1.4 is ridiculous, and slow on everything but Radeon cards, and the only GeForce card that uses it is the FX. There's also no real advantage to it.

[tinyp3nis]

Originally posted by Asher

That's not the problem, though.

The GeForce FX supports 1.4, it just needs to be emulated. 3DMark pings the drivers, determines it has 1.4 support, and runs them. They're far slower than Nvidia's equivalent 1.3. One of the things they got busted for cheating is forcing the GeForce FXs to run it in 1.3, since it makes so much more sense.



The only card it performs better on is the Radeon. Don't let the numbering confuse you, 1.4 is not inherently better than 1.3. They're just different versions that MS okayed. Nvidia made 1.1-1.3 (1.0 never got released), ATI made 1.4. They're just different ways to go about doing pixel shading.

Since ATI cards also support 1.1-1.3, virtually no game has implemented 1.4 since it's rather silly to. Which is why I'm completely dumbfounded why Futuremark made 3DMark03 so heavy on it. Hell, in all but one of the DX9 tests, it still uses 1.4 instead of 2.0! It's like somebody went out of their way to make the Radeon's performance better, but that's just my opinion. Either that, or gross stupidity.


What's most likely is no games, at all, will use 1.4. If it's a DX8 game, it'll use 1.3, if it's a DX9 game, it'll use 2.0.

1.4 is ridiculous, and slow on everything but Radeon cards, and the only GeForce card that uses it is the FX. There's also no real advantage to it.

Now hold on there, 1.3 faster than 1.4? I would love to see a link. Seriously, I haven't heard about this being said not even once. I also thought that 1.4 (like 2.0) is supposed(!!) to be in the card if it is fully directx9 card. It's supposed to be one of the dx9 specifications, not just ATI thing?
Links!

[tinyp3nis]

1.3 is "faster" on Geforce, because they can't do 1.4 well, is that what you are saying?
If so, it's not the same thing as 1.3 actually being faster than 1.4!
Nvidia does have monopoly, and the games will be made so that they work on most+slowest cards.

[Asher]

Originally posted by tinyp3nis
Now hold on there, 1.3 faster than 1.4? I would love to see a link. Seriously, I haven't heard about this being said not even once. I also thought that 1.4 (like 2.0) is supposed(!!) to be in the card if it is fully directx9 card. It's supposed to be one of the dx9 specifications, not just ATI thing?
Links!

1.3 is much faster than 1.4 on the GeForce FX, because the 1.4 is translated into 2.0 code, THEN evalulated by the GeForce FX. Not only is there the extra translation step, there's also the overhead of running 2.0 code to do simple 1.4 things.

The GeForce FX runs 1.1-1.3 natively, 2.0 natively, and 1.4 is emulated.

1.3 is "faster" on Geforce, because they can't do 1.4 well, is that what you are saying?
If so, it's not the same thing as 1.3 actually being faster than 1.4!

Neither is faster than eachother if implemented equally. 1.4 is drastically different than 1.3, one is not better than the other. 1.3 is Nvidia's DX8.1 implementation, 1.4 is ATI's DX8.1 implementation. Nvidia turned their spec in before, so it was labeled 1.3, ATI turned theirs in later, so it got 1.4.

The thing is, the only video cards who support 1.4 natively are Radeons. 1.3 is supported natively by S3, SiS, Nvidia, and ATI. That's why games use 1.3 for DX8, and 2.0 for DX9.

1.4 is a failed addition by ATI, and the only mainstream app who uses it is 3DMark03. And they use it very heavily.

[tinyp3nis]

So in short, do you think Futuremark was lying when they said that 1.1-> 1.3 had no real performance advantage, but 1.1-> 1.4 did have some.

[Asher]

Originally posted by tinyp3nis
So in short, do you think Futuremark was lying when they said that 1.1-> 1.3 had no real performance advantage, but 1.1-> 1.4 did have some.

It has a performance advantage on the Radeon only, which is why I find it strange that ATI sent developers to help with the program, and why Nvidia quit it in protest...

And the reason it has a performance advantage on the Radeon is because the Radeon was designed around using 1.4, and 1.1/1.3 was added at the last minute, but still run natively.

[tinyp3nis]

Now that I can agree with The reason why the better(only little but still) 1.4 wont be used is because Nvidia refuses to use it, FX should have had it in it.

[Asher]

The FX doesn't have it natively, because they didn't think they'd need it. Why waste more transistors when you've already got 130M of them?

No games used 1.4 then, only 2 use them now (and also have native 1.3). 3DMark is a very notable exception...

[tinyp3nis]

2 games that uses it is 2 games more than I knew of What's their names?

[Asher]

Hell if I can remember them.

[Ned]

Originally posted by Fez
I aways played MOHAA.. my favorite game indeed... I never played Ghost recon though.

In Vietcong, you usually have a medic. Or sometimes first aid kits. If you play on easy you can always get healed to 100%.. but if you play on normal or harder.. you will notice the medic only can do so much.. you sustain injury overtime no matter if you get treated by the medic.

That is exactly how it is in Vietcong too.. if you play Vietcong mode.. there is no save. And it is insanely difficult. I tried it. They say in the manual.. that is how it was like.

Also you have to keep your team alive because each member has a specific function.

Well Fez, I just got through Vietcong on easy mode. It was really hard. I can only imagine what it would be like on Vietcong mode. For example, in most cases I would just run up to the VC and blaze away with my automatic weapon. If I were killed with only one hit, this would be impossible.

That last set mission, defending the base, was made a lot harder because I started with the wrong weapon. I think the shotgun is best. I had the M1 Garand Sniper, which I quickly had to throw away in fravor of a shotgun.

I can see how this game has a lot of replayabilty with the choices of weapons on each mission. A very good game, overall.

BTW, the book We Were Soldiers reported how loud the battles were in Vietnam with so many authomatic weapons shooting all at once. I think this game capturef that aspect of the war to a tee. That last battle was really loud.

[Cruddy]

Originally posted by Ned

... that Russia took enormous casualties in assualting the German army in 44-45, suffering far more casualties than it inflicted on the Germans, while at the same time the US Army took very few casualties and inflicted many times that number on the Germans? Was the primary difference in the two theaters the use of airpower by the US? Or was there something more, like bad Russian tactics?)

Actually, many factors. Most important was lack of option to surrender on Eastern Front (you became a casualty anyway).

After reading this thread, and trying the latest NVid driver for my obsolescent TNT2 (it trashed my hard disk ), I can safely say my next upgrade is going to be a Radeon.

And I will be playing Ghost Recon and Vietcong on it (there are lots of the latter in the local exchange game stores - it must be very tough).

[Ned]

Originally posted by Cruddy


Actually, many factors. Most important was lack of option to surrender on Eastern Front (you became a casualty anyway).

After reading this thread, and trying the latest NVid driver for my obsolescent TNT2 (it trashed my hard disk ), I can safely say my next upgrade is going to be a Radeon.

And I will be playing Ghost Recon and Vietcong on it (there are lots of the latter in the local exchange game stores - it must be very tough).

Cruddy, The two games are simply extraordanary.

As to the differences between the Russian and American performance, I can hardly think that it has anything to do with the doggedness of the Germans on the Eastern Front. In the West, we killed a lot more Germans per American. It seems, rather, that the Germans were not surrendering on the Western Front.

The Russians had a lot of airpower as well. So I am beginning to think that the Russians were big on human wave charges.